What does it take to move generative AI from prototype to production - and keep it learning at scale? This session breaks down the ATTD loop (Analyse, Train, Test, Deploy), a framework for deploying AI systems that evolve continuously through real-world usage. Drawing on practical lessons from production environments with millions of user interactions, we’ll explore how to structure feedback pipelines, manage model drift, and ensure reliability in systems that never stop learning. Whether you're working on LLMs, multimodal models, or retrieval-augmented systems, this session offers a grounded perspective on what it really takes to operationalize generative AI.